Gut Microbiota Patterns Associated with Colonization of

Different Clostridium difficile Ribotypes

Jure Skraban1,3, Saso Dzeroski3,4,5, Bernard Zenko3,4, Domen Mongus6, Simon Gangl6, Maja Rupnik1,2,3*
1 Faculty of Medicine, University of Maribor, Maribor, Slovenia, 2 Centre for Microbiology, Institute of Public Health Maribor, Maribor, Slovenia, 3 Centre of Excellence for
Integrated Approaches in Chemistry and Biology of Proteins, Ljubljana, Slovenia, 4 Jozef Stefan Institute, Ljubljana, Slovenia, 5 Jozef Stefan International Postgraduate
School, Ljubljana, Slovenia, 6 Faculty of Electrical Engineering and Computer Science, University of Maribor, Maribor, Slovenia

Abstract
C. difficile infection is associated with disturbed gut microbiota and changes in relative frequencies and abundance of
individual bacterial taxons have been described. In this study we have analysed bacterial, fungal and archaeal microbiota by
denaturing high pressure liquid chromatography (DHPLC) and with machine learning methods in 208 faecal samples from
healthy volunteers and in routine samples with requested C. difficile testing. The latter were further divided according to
stool consistency, C. difficile presence or absence and C. difficile ribotype (027 or non-027). Lower microbiota diversity was a
common trait of all routine samples and not necessarily connected only to C. difficile colonisation. Differences between the
healthy donors and C. difficile positive routine samples were detected in bacterial, fungal and archaeal components.
Bifidobacterium longum was the single most important species associated with C. difficile negative samples. However, by
machine learning approaches we have identified patterns of microbiota composition predictive for C. difficile colonization.
Those patterns also differed between samples with C. difficile ribotype 027 and other C. difficile ribotypes. The results
indicate that not only the presence of a single species/group is important but that certain combinations of gut microbes are
associated with C. difficile carriage and that some ribotypes (027) might be associated with more disturbed microbiota than
the others.

Citation: Skraban J, Dzeroski S, Zenko B, Mongus D, Gangl S, et al. (2013) Gut Microbiota Patterns Associated with Colonization of Different Clostridium difficile
Ribotypes. PLoS ONE 8(2): e58005. doi:10.1371/journal.pone.0058005
Editor: Michel R. Popoff, Institute Pasteur, France
Received November 19, 2012; Accepted January 29, 2013; Published February 28, 2013
Copyright: 2013 Skraban et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits
unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: This work was supported by Slovenian Research Agency grants 1000-10-310160 and J3-4298 and Operation no. OP13.1.1.2.02.0005 financed by the
European Regional Development Fund and the Slovenian Ministry of Higher Education, Science and Technology. The funders had no role in study design, data
collection and analysis, decision to publish, or preparation of the manuscript.
Competing Interests: The authors have declared that no competing interests exist.
* E-mail: maja.rupnik@zzv-mb.si

Introduction C. difficile 027 ribotype positive subjects (n = 2) with diarrhoea at

Clostridium difficile infection (CDI) is recognised as the leading
cause of nosocomial intestinal infections, mainly in elderly
hospitalised patients 1]. The onset of CDI has been linked to
the use of broad-spectrum antibiotics, historically especially with
clindamycin 2,3]. This implicated very early that the healthy gut
microbiota has an important role in disease development. The
importance of gut microbes is also supported by the high success
rate (around 90%) of biotherapy (faecal transplantations) used in
patients with severe cases of recurring CDI 46].
Several studies on gut microbiota changes associated with C.
difficile colonization or infection have been published recently.
Mice treated with clindamycin loose up to 90% of bacterial taxons
in the cecum and are rapidly colonized with C. difficile 7]. Other
molecular studies on the changes of gut microbiota during CDI
have used16S rRNA clone libraries and showed lower bacterial
microbiota diversity in patients with a recurring CDI 8]. The
banding patterns on gradient gels connected specific bacterial
groups to C. difficile colonisation in adults 9] and infants 10].
Another study used 454 sequencing to compare the composition of
gut microbiota of C. difficile positive (20 asymptomatic and 2
diarrhoeal samples) and negative subjects (252 samples). They
found no differences in the composition of faecal microbiota in
asymptomatic C. difficile carries and healthy subjects. However, in
the time of sampling faecal microbiota had a lower overall
diversity at the genus level 11].
As described, different molecular approaches (including dena-
turing gradient gel electrophoresis (DGGE), 16S rRNA clone
libraries and metagenomics) were used in studies on C. difficile and
changes in microbiota. Here we have applied a simple and high-
throughput DHPLC method (DHPLC - denaturing high pressure
liquid chromatography) which separates 16S rDNA amplicons
based on fragment size and sequence to analyse human faecal
microbiota in the C. difficile positive and negative routine samples.
It has been shown previously, that DHPLC offers a level of
analysis, which is similar to gradient gels (DGGE) in terms of
sensitivity and resolution, with the advantage of high automatisa-
tion and repeatability 12,13]. It has been used successfully in
community profiling of microbiota of the urogenital tract 14],
detecting changes in gut microbiota of patients during antimicro-
bial therapy 12], following the development of gut microbiota in
infants 15], and in profiling various other complex communities
from fermenter sludge, compost or soil 13].
Studies on the association of gut microbiota and C. difficile
mostly describe the bacterial microbiota and very rarely fungi 16],
while no studies have included archaea. With DHPLC we have
analysed the bacterial, but also the fungal and archaeal gut
microbiota in human faecal samples with requested C. difficile

PLOS ONE | www.plosone.org 1 February 2013 | Volume 8 | Issue 2 | e58005


testing. With machine learning analysis of the collected data, we
were able to show that certain patterns of microbiota composition
are associated with C. difficile colonization and in particular with
the PCR ribotype 027.
Materials and Methods
Collecting and storing of the faecal samples
Faecal samples with requested C. difficile testing (n = 171) were
obtained from two different routine laboratories. These samples
were routinely collected in plastic containers and have been stored
at 4uC and shipped to the study site within 48 h, where they have
been immediately frozen at 280uC until further processing.
According to the results of the routine C. difficile testing, 105
faecal samples were C. difficile positive and 66 were C. difficile
negative. A stool sample was described as non-diarrhoeal or
diarrhoeal (has taken the shape of the container). No further
clinical information was obtained for the samples.
Additional 37 samples were collected from healthy donors and
stored as described above. None of the samples from healthy
donors was diarrhoeic. The age of the donors ranged from 14 to
72 years. The gender distribution was 22 females and 15 males.
Ethics Statement
The National Medical Ethics Committee of the Republic of
Slovenia (NMEC) has approved the study (number 126/05/12).
NMEC has also approved that written or verbal consent is not
needed for routine samples and that only verbal informed consent
is needed for healthy volunteers. Documenting was not requested.
In the case of minor/children participants informed consent was
obtained from the parents.
Detection, isolation and ribotyping of C. difficile from
faecal samples
In each of the two routine laboratories, C. difficile was detected
by different methods. One laboratory performed the VIDAS
ToxA/ToxB test and culturing on CLO selective plates (BioMer-
ieux) after ethanol shock. The second laboratory used the
molecular Cepheid Xpert C. difficile assay. From the samples
tagged as C. difficile positive by the Cepheid Xpert C. difficile assay,
C. difficile was isolated by enrichment in Oxoid C. difficile broth with
Oxoid supplement SR0096E, lysozyme (5 mg/l) and sodium
cholate (1%) as described previously 17]. The same enrichment
was also used to test the presence of C. difficile in the faecal samples
of the healthy volunteers.
All isolates were confirmed as C. difficile by the amplification of
cdd3, located downstream from the pathogenicity locus (PaLoc) 17]
and ribotyped with primers described by Bidet 18]. The profiles
were analysed with the BioNumerics software. All isolates were
also toxigenic as determined by toxinotyping (data not shown) 19].
Isolation of the total DNA from faecal samples and
amplification of the marker genes
The extraction of DNA was performed with the DNA Stool
Mini Kit (Qiagen, Germany) after mechanical disruption with the
SeptiFast Lyse Kit (Roche) on MagNA Lyser (speed 7000, 70 s;
Roche). The DNA concentration and purity were checked with
Nanodrop and 60 ng of faecal DNA was used for PCR
amplification of bacterial, archaeal and fungal marker genes.
Primers and cycling conditions described by Domann 14] were
used for amplification of the variable region V6V8 of bacterial
16S rRNA genes. The correct amplicon size (470 bp) was checked
by agarose gel electrophoresis. Heteroduplex DNA molecules were
PLOS ONE | www.plosone.org
Gut Microbiota Patterns and C. difficile
reduced by reconditioning as described 20]. Before loading onto
the DHPLC column, the PCR products were purified with the
PCR Purification Kit (Qiagen).
For the amplification of archaeal 16S rRNA genes, we have
targeted the variable region V3 for which we used a modified
forward primer PARCH340F 59- CCCTACGGGGTGCAGCAG
-39 previously described by Ovreas 21] and designed a new reverse
primer A780R 59- TACCCGGGTATCTAATCCGGT -39 which
anneals to a conserved region about 420 bp downstream. The
cycling conditions were 95uC for 7 min, followed by 35 cycles
using 95uC for 40 s, 63uC for 30 s, 72uC for 30 s and the final
extension at 72uC for 5 min. We have checked the amplicon size,
reconditioning and purification as described above.
The fungal internal transcribed spacer region 2 (ITS2) was
amplified using a nested PCR. For the first reaction, we have used
the primer pair nu-SSU-0817F 59-TTAGCATGGAATAATR-
RAATAGGA-39 22] and ITS4R-59 TCCTCCGCTTATTGA-
TATGC-39 23]. The cycling conditions were 95uC for 2 min,
followed by 14 cycles using 95uC for 30 s, 61.3uC for 30 s with a
0.5uC decrement for each following cycle and 72uC for 60 s. This
was followed by 19 cycles at 95uC for 30 s, 54.3uC for 30 s and
72uC for 60 s and the final extension at 72uC for 5 min. The
correct size of the amplicons (14002000 bp) was checked by
agarose gel electrophoresis. Before the second PCR reaction, we
have degraded any remaining primers from the first PCR reaction.
For the degradation, the Shrimp Alkaline Phosphatase and
Exonuclease I (both from Fermentas) were used. The reaction
mixture of the final volume of 6.5 ml contained 0.15 U/ ml of the
Phosphatase, 1.5 U/ ml of the Exonuclease I and 5 ml of the PCR
product from the first reaction. The mixture was incubated at
37uC for 15 min and then the enzymes deactivated (85uC,
15 min). Five ml were then used as the template for the second
PCR reaction using the primer ITS86F 59-GTGAATCATC-
GAATCTTTGAAC-39 24] and the ITS4R primer described
above. The cycling conditions were 95uC for 2 min, followed by
25 cycles at 95uC for 30 s, 54.3uC for 30 s and 72uC for 40 s and
the final extension at 72uC for 5 min. The right size of the
amplicons (approximately 200400 bp) was again checked on
agarose gel and the products purified as described above.
Separation of the amplified DNA with DHPLC
To separate the PCR products, we have used denaturing high
pressure liquid chromatography (DHPLC; WAVE Microbial
Analysis System; Transgenomics, USA). The amplicon volumes
loaded on the column were 20 ml for bacteria, 15 ml for archaea
and 10 ml for fungi. Conditions for separating the bacterial and
fungal amplicons were described previously by Domann 14] and
Goldenberg 25], respectively. For archaea, we have optimised the
separation conditions which are given in the Supplementary table
S1. A UV detector was used in DNA detection and the results
were analysed with the Navigator Software version 2.2.0 (Build
25), the peak analysis parameters are given in the Supplementary
table S1.
Normalization of retention times and assigning the
DHPLC peaks to microbial groups
Before analysis, the retention times of the DHPLC peaks were
normalized. A marker designed as a G - C rich oligonucleotide,
which eluted after the sample, was loaded onto the column
together with the sample. The retention times of the peaks were
then adjusted according to the equation
NRT~RTz(RT  (1-MRT=X )), where NRT stands for the
normalized retention time, RT for the retention time, MRT for
2 February 2013 | Volume 8 | Issue 2 | e58005

the retention time of the marker and X for an arbitrarily chosen
point on the retention time scale.
Identification of the microbial groups
The retention time scale of the DHPLC profiles was divided
into 0.2 min intervals. The representative peaks within each
0.2 min interval were collected, sequenced and all peaks within the
given 0.2 min interval were then assigned to a bacterial group as
described below.
The DHPLC peak fractions were collected form well separated
and overlapping peaks and sent for sequencing (Eurofins MWG
Operon, Germany). In case of partially overlapping peaks, the
collected DHPLC peak fractions sometimes resulted in mixed
sequence. In this case the corresponding PCR products were
cloned into pGEMH-T Easy Vector System (Promega, Germany)
following the manufacturers instructions. Ten colonies from each
cloned fraction were collected, 16S rRNA gene fragments re-
amplified (as above), purified (as above) and screened by DHPLC.
Those, which differed in retention times, were selected and
sequenced (Eurofins MWG Operon, Germany). The sequences
were compared to the databases provided by the National Centre
for Biotechnology Information (NCBI) and Ribosomal Database
Project-Release 10 and the DHPLC peak fractions named
accordingly. All 0.2 min intervals were named after the organisms
identified in the fractions belonging to a particular interval. If
more than one organism was identified in the DHPLC fractions
taken within a particular 0.2 min interval, that interval was named
after all the identified organisms. The named time intervals were
from now on referred to as microbial groups.
Statistical analysis
The 208 faecal samples were divided into seven sets of samples,
based on the following criteria: healthy donors or routine samples,
diarrhoea or formed stool, the presence or absence of C. difficile,
and 027 ribotype or other ribotypes. The names of the sets and the
number of samples in each set are given in Tables 1, 2 and 3.
The results were analysed by looking at the differences between
the sets of samples in terms of presence or absence of microbial
groups or by looking at the percentage of peak area within the
Table 1. Comparison of routine C. difficile negative samples (set of samples with formed stool and set with unformed stool) vs.
healthy donors by logistic regression model.
a
Sample set 1 vs. Sample set 2
Peptostreptococcaceae
Ruminococcus bromii
Bifidobacterium longum
Enterobacteriaceae 3
Faecalibacterium sp.
Bacteroides uniformis
Prevotella sp.
Alistipes sp.
Escherichia/Shigella sp.
Bacteroides vulgatus
a - the compared sample sets and the number of samples in each set is specified.
NEG-F: C. difficile negative/formed stool; NEG-D: C. difficile negative/diarrhoea; HEALTHY: healthy donors.
Results are presented as odds ratios for individual microbial groups between different sets of samples. An odds ratio below 1 indicates that this particular microbial
group is less frequent in sample set 1 as compared to sample set 2. The odds ratios are given in the table only for the p-values (shown in parentheses) below 0.05 for the
compared sample sets.
doi:10.1371/journal.pone.0058005.t001
PLOS ONE | www.plosone.org
Gut Microbiota Patterns and C. difficile
total analysed chromatogram area which was computed by the
Navigator Software version 2.2.0 (Build 25) using parameters as
described in the Supplementary table S1.
The difference in the number of organism groups was analysed
with ANOVA. The equality of the variances in each group was
first verified with the Levene test and the pre -planned post-hoc
comparisons were carried out with contrast analysis. Since the
observed contrasts were not orthogonal, the p-values were
adjusted with the Holm method to control the type I error.
Diversity of microbiota was assessed with Reciprocal Simpsons
index of diversity and the differences between the compared sets of
samples evaluated with Student T-test.
Logistic regression was used to analyse the presence or absence
of individual microbial groups. The dependency between the
dependent variable (1- individual has an organism group, 0 -
individual does not have an organism group) and a microbial
group was first tested with the chi - square test. After logistic
regression, we used a contrast analysis to carry out the pre-planned
comparisons between the sets of samples.
Principal component analysis
Principal Component Analysis (PCA) with respect to Instru-
mental Variables was performed with R programme (version
2.15.1) and applications ade4 (version 1.50) and ade4TkGUI
(version 0.25). We used the data matrix based on the presence of
the peak and percent area of the peaks (representing an individual
microbial group) from the DHPLC chromatograms to calculate
the spatial coordinates of each faecal sample. We also calculated
the centre of gravity and dispersion for each set of samples.
Statistical significance of PCA clustering based on microbiota
profiles was assessed using a Wilcoxon signed-rank test.
Machine learning analysis
We further analysed the differences between the sets of samples
in terms of presence or absence of microbial groups by using
machine learning methods for the induction of decision trees 26],
more specifically classification trees 27]. In particular, we used the
WEKA J48 implementation 28] of the C4.5 algorithm 26]. The
default parameter settings for J48 were applied, which include a
NEG-F (29) vs. Healthy (37) NEG-D (37) vs. Healthy (37)
0.21 (0.0028)
0.11 (0.0454)
0.27 (0.0288)
0.25 (0.0088) 0.25 (0.0056)
0.08 (0.0201)
0.29 (0.0161)
0.23 (0.0256) 0.18 (0.007)
0.2 (0.0133) 0.14 (0.0014)
0.17 (0.0012) 0.21 (0.0028)
0.11 (0.0059) 0.3 (0.0331)
3 February 2013 | Volume 8 | Issue 2 | e58005
 Gut Microbiota Patterns and C. difficile

Table 2. Comparison of different sets of routine C. difficile positive samples vs. healthy donors by logistic regression model.

OTHER-F (17) vs. OTHER-D (31) vs. 027-F (16) vs. 027-D (41) vs.
a
Sample set 1 vs. Sample set 2 Healthy (37) Healthy (37) Healthy (37) Healthy (37)

Peptostreptococcaceae 0.25 (0.0285)

Ruminococcus bromii 0.05 (0.0069)
Bifidobacterium longum 0.18 (0.0341) 0.25 (0.0197) 0.09 (0.0247) 0.03 (0.0013)
Enterobacteriaceae 1 0.07 (0.018) 0.06 (0.0149) 0.1 (0.0323)
Bacteroides sp. 1 0.3 (0.0461) 0.35 (0.0385) 0.19 (0.011) 0.22 (0.0019)
Enterobacteriaceae 2 0.26 (0.0116) 0.15 (0.0036) 0.13 (0.0001)
Enterobacteriaceae 3 0.2 (0.0021) 0.07 (0.0013) 0.08 (0.0000)
Bacteroides sp. 2 0.28 (0.0326) 0.1 (0.0322) 0.12 (0.0017)
Faecalibacterium sp. 0.04 (0.0042) 0.1 (0.0323)
Bacteroides uniformis 0.35 (0.0403) 0.12 (0.0106) 0.27 (0.0084)
Prevotella sp. 0.17 (0.0153) 0.22 (0.0198) 0.12 (0.0038) 0.15 (0.0025)
Alistipes sp. 0.11 (0.0019) 0.19 (0.0105) 0.14 (0.0014)
Escherichia/Shigella sp. 0.13 (0.0022) 0.09 (0.0000) 0.19 (0.0105) 0.15 (0.0002)
Bacteroides vulgatus 0.1 (0.0044) 0.12 (0.0017)
Candida albicans 3.62 (0.0115)
Candida glabrata 10.45 (0.0301)
Methanobrevibacter smithii 0.23 (0.0189) 0.20 (0.0025) 0.25 (0.0041)

a - the compared sample sets and the number of samples in each set is specified.
OTHER-F: C. difficile non 027 ribotype/formed stool; OTHER-D: C. difficile non 027 ribotype/diarrhoea; 027-F: C. difficile 027 ribotype/formed stool; 027-D: C. difficile 027
ribotype/diarrhoea; HEALTHY: healthy donors.
Results are presented as odds ratios for individual microbial groups between different sets of samples. An odds ratio below 1 indicates that this particular microbial
group is less frequent in sample set 1 as compared to sample set 2. The odds ratios are given in the table only for the p-values (shown in parentheses) below 0.05 for the
compared sample sets.
doi:10.1371/journal.pone.0058005.t002

minimum of two examples per leaf for pre-pruning and a
confidence level of 0.25 for post-pruning.
The dependent (class) variable in this analysis was the ribotype
of C. difficile detected in a sample (none, 027, other). The
independent variables (attributes) were the indicators of presence/
absence of each of the 34 microbial groups, as well as an indicator
of the sample category (healthy donor/routine sample).
We next reversed the direction of analysis and analysed the
overall microbiota composition depending on the detected
ribotype of C. difficile and other sample properties. To this end,
we used predictive clustering trees - PCTs 29] for multi-target
classification, as implemented in the Clus data mining suite
(http://clus.sourceforge.net). Each of the indicators of presence/
absence of the 34 microbial groups was considered as a target
(class) variable, all of them to be predicted simultaneously by a

Table 3. Comparisons of all routine C. difficile positive vs. all routine C. difficile negative samples by logistic regression model.

OTHER-D (31) vs. 027-D (41) vs. 027-D (41)/OTHER-D (31) vs.
a
Sample set 1 vs. Sample set 2 NEG-D (37) NEG-D (37) NEG-D (37)

Streptococcus sp./Enterococcus sp. 2 3.7 (0.0471) 3.23 (0.0461)

Peptostreptococcaceae 2.9 (0.0259) 2.63 (0.0232)
Ruminococcus bromii 4.8 (0.0491) 3.33 (0.0453)
Bifidobacterium longum 0.04 (0.0031) 0.12 (0.0012)
Enterobacteriaceae 1 0.1 (0.037)
Bacteroides sp. 1 0.39 (0.0465)
Enterobacteriaceae 2 0.35 (0.0289)
Enterobacteriaceae 3 0.33 (0.0498)
Methanobrevibacter smithii 0.39 (0.0491) 0.31 (0.0268) 0.35 (0.0138)

a - the compared sample sets and the number of samples in each set is specified.
NEG-D: C. difficile negative/diarrhoea; OTHER-D: C. difficile non 027 ribotype/diarrhoea; 027-D: C. difficile 027 ribotype/diarrhoea.
Results are presented as odds ratios for individual microbial groups between different sets of samples. An odds ratio below 1 indicates that this particular microbial
group is less frequent in sample set 1 as compared to sample set 2. The odds ratios are given in the table only for the p-values (shown in parentheses) below 0.05 for the
compared sample sets.
doi:10.1371/journal.pone.0058005.t003

PLOS ONE | www.plosone.org 4 February 2013 | Volume 8 | Issue 2 | e58005

 Gut Microbiota Patterns and C. difficile

single tree. As independent variables (attributes), we considered the
sample category (healthy donor/routine sample), the consistency
of the stool (formed/diarrhoea) and the ribotype of C. difficile
detected in a sample (none, 027, other).
Results
The aim of this study was to compare the gut microbiota in
faecal samples from healthy donors and from routine samples
negative or positive for C. difficile and to detect possible differences
in dominant microbial groups (bacteria, archaea, fungi).
We have analysed 208 faecal samples, of which 171 were
routine samples and 37 were from healthy volunteers. All faecal
samples from healthy volunteers were C. difficile negative. Of the
171 routine samples, 105 were C. difficile positive and 66 were C.
difficile negative. From all 105 positive faecal samples C. difficile was
isolated and strains were assigned to 22 different C. difficile PCR
ribotypes. The five most frequent ribotypes were 027 (57), 014/
020 (6), 081 (6), 002 (5) and 023 (5).
DHPLC analysis of archaeal, fungal and bacterial
microbiota
The representative DHPLC profiles in Fig. 1 show different
complexity of archaeal, fungal and bacterial microbiota in human
faecal samples. This section will only briefly list archaeal, fungal
and bacterial groups detected by the DHPLC method, while the
differences in faecal microbiota between samples (healthy, C.
difficile positive and negative samples) will be described in more
detail in further sections.
The archaeal microbiota was simple in all 208 analysed samples
and was composed maximally of two methanogenic archaea,
Methanosphaera stadtmanae and Methanobrevibacter smithii (Fig. 1a).
The fungal microbiota was more diverse (Fig 1b, Fig. 2). We
have found nine different fungi altogether. The most frequent
among them were Saccharomyces cerevisiae (in 57% of samples) and
Candida albicans (in 27% of samples). Candida glabrata, Candida diversa,
Clavispora lusitaniae, Pichia burtonii, Lemoniera sp., Eurotium sp. and
Scleroderma sp. were found only sporadically.
Bacterial microbiota was most complex (Fig. 1c, Fig. 3).
Altogether we have identified 23 different bacterial groups. Some
chromatographic peaks were assigned to more than one genus
because they either contained mixed sequences, or because the
short sequences used in the DHPLC analysis did not allow
distinction between them (e.g. Clostridium difficile/Sporacetigenium sp.;
Citrobacter freundii/Enterobacter sp.; Streptococcus sp./Enterococcus sp. 1
and 2). Bacterial groups that were identified to the same genus or
family but had different retention times were designated by a
number (Enterobacteriaceae 1, 2 and 3; Bacteroides sp. 1 and 2,
Streptococcus sp./Enterococcus sp. 1 and 2). Because short fragments
are used in the DHPLC analysis further identification of particular
amplified fraction was not possible. The remaining identified
groups were Clostridium sp., Peptostreptococcaceae, Blautia sp., Rumino-
coccus bromii, Ruminococcus sp., Faecalibacterium sp., Klebsiella pneumonia;
Escherichia/Shigella sp., Prevotella sp., Alistipes sp., Bacteroides uniformis,
Bacteroides vulgatus, Bifidobacterium longum and Bifidobacterium sp..

Figure 1. Representative DHPLC profiles of faecal microbiota. The profiles show archaeal (a), fungal (b) and bacterial (c) faecal microbiota.
The faecal profiles of healthy volunteers are in the upper row and the profiles from the diarrhoeal patients colonised by Clostridium difficile 027
ribotypes are in the lower row. The selected profiles show typical chromatograms of the three studied microbial groups. The chromatograms show
increasing complexities from archaeal to fungal and bacterial microbiota, respectively.
doi:10.1371/journal.pone.0058005.g001

PLOS ONE | www.plosone.org 5 February 2013 | Volume 8 | Issue 2 | e58005

 Gut Microbiota Patterns and C. difficile

Figure 2. Fungal DHPLC profiles of all seven sets of samples. The samples were distributed into sets according to consistency, source
(healthy/routine) and C. difficile presence. The peaks with different retention times (x - axis) represents different fungal groups.Fungal microbiota in
healthy volunteers and C. difficile positive samples differs in the composition (position of the peaks) but not in diversity (number of the peaks).
doi:10.1371/journal.pone.0058005.g002

DHPLC profiles and principal component analysis
All 208 faecal samples were partitioned into seven sets
according to origin (routine sample/healthy), appearance (diar-
rhoea/formed) and C. difficile colonization status (negative/
ribotype 027/any other ribotype). We have named the sample
sets accordingly (diarrhoeal/C. difficile negative samples; diarrhoe-
al/C. difficile ribotype 027 positive samples; diarrhoeal/C. difficile
other ribotypes positive samples; formed stool/C. difficile negative
samples; formed stool/C. difficile ribotype 027 positive samples;
formed stool/C. difficile other ribotypes positive samples; healthy
donors) (Fig. 2 and 3). Because of the large number of samples
colonised with the 027 ribotype (57 out of 105 colonised samples),
we were able to separate those from other ribotypes and treat it as
a separate set of samples.
To compare the faecal profiles of the dominant microbiota in
the seven sets of samples, we first used PCA to calculate the spatial
coordinates of each sample (Fig. 4). All sets of samples showed a
large dispersion, particularly samples sent for routine C. difficile
testing (Fig. 4a). Although the centres of gravity were different for
samples from healthy persons and for routine samples, they also
substantially overlapped (Fig. 4a). The result was similar when we

PLOS ONE | www.plosone.org 6 February 2013 | Volume 8 | Issue 2 | e58005

 Gut Microbiota Patterns and C. difficile

Figure 3. Bacterial DHPLC profiles of all seven sets of samples. Bacterial microbiota in healthy individuals is composed of a large number of
species, hence the peaks are numerous but the area below each peak is small. The presence of C. difficile correlates with decreased diversity reflected
by a lower number of peaks which are, however, more prominent.
doi:10.1371/journal.pone.0058005.g003

compared the samples based on appearance (diarrhoea/formed)
and C. difficile colonisation status (Fig. 4b, c and d).
Regardless of the overlap between the sets of samples, Wilcoxon
signed-rank test performed on each pair of the sets from Fig. 4
(healthy donors or routine samples, diarrhoea or formed stool, the
presence or absence of C. difficile, and 027 ribotype or other
ribotypes) show a clear distinction between them. In all the cases,
location shift greater than 0.5 can be shown with 95% confidence,
while the probability of sets being different is larger than 99.9%.
This confirms that the partition of the samples into the sets at hand
was appropriate and we kept this grouping to further analyse the
differences in microbial groups using statistical analyses and
machine learning methods.
Comparison of microbiota present in routine samples
and in samples from healthy volunteers
Within each of the seven sample sets, we have determined the
average number of bacterial or fungal groups (Fig. 5). The
difference in the average number of present bacterial groups
between the healthy volunteers and each of the six different
routine sample sets was statistically significant (P,0.0001). The
comparison of the average number of bacterial and fungal groups

PLOS ONE | www.plosone.org 7 February 2013 | Volume 8 | Issue 2 | e58005

 Gut Microbiota Patterns and C. difficile

Figure 4. Principal component analysis of DHPLC profiles of dominant bacterial, fungal and archaeal species. Individual samples
(represented by black dots) were clustered into sets according to C. difficile presence and type, origin and consistency (elipses) and the centre of
gravity computed for each set of samples. All sets of samples show large dispersion and overlap (elipses). Figure 4a compares the samples sent for
routine C.difficile testing (pat) and healthy donors (hea). Particularly the former set shows a great dispersion. Figures 4b, c and d compare different
sets of samples from the patients. Figure 4b shows comparison of six different sets of routine samples; a greater shift in the centre of gravity between
the formed and the diarrhoeal C. difficile negative samples (FNP/DNP), compared to the C. difficile positive samples (FOS/DOS, F027/D027). Fig 4c
shows only the subset of different routine diarrheic samples. Fig 4d shows only the subset of different routine formed samples DNP - diarrhoeal/C.
difficile negative samples, D27 - diarrhoeal/C. difficile ribotype 027 positive samples, DOSdiarrhoeal/C. difficile other ribotypes positive samples, FNP -
formed stool/C. difficile negative samples, F27 - formed stool/C. difficile ribotype 027 positive samples, FOS - formed stool/C. difficile other ribotypes
positive samples, pat - samples sent for routine C. difficile testing, hea - healthy donors).
doi:10.1371/journal.pone.0058005.g004

shows that the healthy donors had the highest number of bacterial
groups (in average 16.54) while the highest average among the
routine samples 12.69 was the one for formed/C. difficile negative
stools. The lowest average number of bacterial groups was
observed in the samples colonised with C.difficile ribotype 027
(9.69 and 11.10 for formed and diarrhoeal stools respectively). The
fungal microbiota showed no difference in the average number of
fungi between the healthy donors and routine samples (Fig. 5).
Simpsons reciprocal index of diversity showed similar differences
between the sets of samples with healthy donors having

PLOS ONE | www.plosone.org 8 February 2013 | Volume 8 | Issue 2 | e58005

 Gut Microbiota Patterns and C. difficile

significantly higher microbiota diversity (10.03) than routine Table 4. Differences in microbial groups between healthy
samples (from 4.47 to 8.38). In addition, the samples colonised donors and routine samples as detected by the PCT model.
with the ribotype 027 had significantly lower diversity (4.47 and
5.37, formed and diarrhoeal samples respectively) than samples
colonised with other C. difficile ribotypes (6.25 and 8.38, formed Organism group Healthy donor Routine samples
and diarrhoeal samples respectively). The diversity indexes and P-
Enterobacteriaceae 2 1 0
values are given in Supplementary table S2.
After assessing differences between the average numbers of Enterobacteriaceae 3 1 0
microbial groups, we have further analysed the differences on the Escherichia/Shigella sp. 1 0
level of specific microbial taxons with PCT (predictive clustering Bacteroides sp.1 1 0
trees) analysis and with logistic regression. Several PCTs were Bacteroides uniformis 1 0
built, relating microbiota composition to different combinations of
independent variables. Here we present the trees built using the An entry of 1 means that the organism group (row) was present and an entry of
0 means that the organism group was absent in the majority of samples of the
sample category only (Table 4) and the ribotype of C. difficile only
considered type (column).
(Table 5). The PCT models take into account only microbial doi:10.1371/journal.pone.0058005.t004
groups present at high frequencies (above 50% in at least one
sample set). We highlight the differences between healthy donors
between the healthy donors and C. difficile positive routine samples
and routine samples (Table 4) and between routine samples with
were detected in bacterial, fungal and archaeal components
different C. difficile colonisation/ribotype status (Table 5), and only
(Table 2).
list the microbial groups that are present differentially (i.e., present
in the majority of one set of samples and absent from the majority
of the other set of samples).. When healthy donors were compared
Changes of microbiota associated with Clostridium
to all routine samples (not distributed into sets regarding stool difficile presence in the faecal sample
consistency and C. difficile presence) by PCT, five microbial groups As described before the average number of bacterial and fungal
were identified to be more frequent in healthy donors, including groups was not significantly different between the routine samples
different Enterobacteriaceae and Bacteroides sp. (Table 4). with a different C. difficile colonisation status and ranged from 9.69
Subsequently, all microbial groups were analysed by logistic bacterial groups in diarrhoeal/C. difficile ribotype 027 positive and
regression to calculate the odds ratios for individual microbial to 12.69 bacterial groups in formed/C. difficile negative samples,
groups when comparing sets of samples one to another (Tables 1, 2 respectively, and (on average) from 1.5 to 2.0 fungal groups per
and 3). In logistic regression, the routine samples were also further sample (Fig. 5).
distributed into sets (C. difficile negative, 027 C. difficile positive, When different subsets within the routine samples were
other than 027 C. difficile positive). Healthy donor microbiota compared by the PCT model, which takes into account only
differed from C. difficile negative routine samples mainly in microbial groups present at high frequencies (above 50% in at least
bacterial composition (Table 1). On the other hand, differences one sample set), we identified five microbial groups characteris-

Figure 5. Average numbers of bacterial and fungal groups in different sets of faecal samples. The average numbers of present bacterial
and fungal groups within each sample set are presented as means with confidence bars plot. 027/F: C. difficile 027 ribotype/formed stool; 027/D: C.
difficile 027 ribotype/diarrhoea; NEG/F: C. difficile negative/formed stool; NEG/D: C. difficile negative/diarrhoea; OTHER/F: C. difficile non 027 ribotype/
formed stool; OTHER/D: C. difficile non 027 ribotype; HEALTHY: healthy donors.
doi:10.1371/journal.pone.0058005.g005

PLOS ONE | www.plosone.org 9 February 2013 | Volume 8 | Issue 2 | e58005

 Gut Microbiota Patterns and C. difficile

Table 5. Differences in microbial groups present between different types of C. difficile colonisation as predicted by the PCT model.

Organism group C. difficile negative C. difficile non 027 ribotype C. difficile 027 ribotype

Enterobacteriaceae 2 1 1 0
Bacteroides sp. 1 1 0 0
Escherichia/Shigella sp. 1 0 0
Saccharomyces cerevisiae 1 1 0
Methanobrevibacter smithii 1 0 0

An entry of 1 means that the organism group (row) was present and an entry of 0 means that the organism group was absent in the majority of samples of the
considered type (column)
doi:10.1371/journal.pone.0058005.t005

tically present or absent (the majority of) in different groups of
routine samples (Table 5). All five were predicted to be present in
C. difficile negative routine samples and only two (Enterobacteriaceae
group and S. cerevisiae) in routine samples colonized with C. difficile
ribotype other than 027.
When comparing C. difficile negative or positive samples to
healthy volunteers, bacteria and archaea were less frequent in
routine samples than in healthy donors, while some fungi were
more frequent in C. difficile positive samples (Tables 1 and 2).
When comparing diarrhoeal C. difficile negative samples to
diarrhoeal C. difficile positive samples (regardless of the ribotype)
significant differences were found in four bacterial groups
(Peptostreptococcaceae, Ruminococcus bromii, Streptococcus/Enterococcus sp.
2 and Bifidobacterium longum) and one archaeal group (Methano-
brevibacter smithii) (Table 3, last column). Among them, Bifidobacter-
ium longum and Methanobrevibacter smithii showed lower frequencies in
the C. difficile positive samples. However, these differences were
mainly due to the diarrhoeal 027 positive C. difficile samples as it
will be described below.
Archaea (Methanobrevibacter smithii/Methanosphaera stadtmanae) were
associated with healthy donors (detected in 65% (M. smithii)/22%
(M. stadtmanae) of samples) and C. difficile negative samples (52%/
21% for formed stool and 51%/19% for diarrhoea). C. difficile
positive samples (especially those colonized with non - 027
ribotype) were characterised by lower frequencies of archaea.
Frequencies of M. smithii/M. stadtmanae in 027 positive C. difficile
samples were (38%/19% for formed stool and 32%/17% for
diarrhoea) and in non - 027 positive samples (29%/12% for
formed stool and 26%/10% for diarrhoea). The diarrhoeal non -
027 positive C. difficile samples and diarrhoeal C. difficile negative
samples significantly differed to each other only in Methanobrevi-
bacter smithii (Table 3, first column).
A decision tree was constructed by the J48 machine learning
approach, based on the presence or absence of microbial groups,
to predict the outcome regarding C. difficile colonisation. The
resulting decision tree splits the samples by category first. We only
discuss the part that refers to the routine samples, shown in
Figure 6. This decision tree was built automatically from the data
on the collected samples. Each internal node is labeled with a test
on the presence/absence of a microbial group, while each terminal
node (leaf) is labeled with the most frequent class/outcome (C.
difficile 027/C.difficile other ribotypes/C. difficile negative). The
numbers in parentheses provide an indication of the accuracy of
the prediction made by the leaf (total number of samples in the
leaf/incorrectly classified samples). The tree therefore summarizes
the data on the samples and the results of their analysis, relating
microbial group composition to the outcome.
The decision tree can also be used as a predictive model, i.e., to
make predictions regarding the C. difficile ribotype of new samples
based on their microbial groups. In this case, its accuracy would
have to be estimated on unseen data (and not on the training data
as done here) by using procedures such as cross-validation.
The tree identifies several different combinations of microbial
groups associated with the same outcome. The presence as well as
absence of microbial groups influenced the outcome importantly.
The microbial groups identified by the model to be most related to
the outcome were bacteria, but also fungi and archaea (Fig. 6).
The most strongly related microbial group was Bifidobacterium
longum the presence of which was linked to the C. difficile negative
outcome in 18 out of 27 cases. The presence of Prevotella sp. was
associated with both, C. difficile positive and negative status,
depending on the presence or absence of other microorganisms. It
was associated with C. difficile absence when found together with
Enterobacteriaceae 1 and 2. This was also the pattern with the best
classification accuracy, misclassifying only 2 out of 18 samples. On
the other hand, Prevotella sp. was associated with C. difficile presence
when found together with Enterobacteriaceae 3 and Streptococcus sp./
Enterococcus sp. 1.
Specific changes in microbiota in the patients colonised
with the C. difficile ribotype 027
Because a large number of C. difficile positive routine samples
had ribotype 027 (57/105) we could study those samples as a
separate set. We found most differences (8 bacterial groups)
between the diarrhoeal C. difficile negative and diarrhoeal ribotype
027 positive samples (Table 3). Three bacterial groups (Peptos-
treptococcaceae, Ruminococcus bromii and Streptococcus/Enterococcus sp. 2)
were more frequent and five bacterial groups (Enterobacteriaceae 1, 2,
3; Bacteroides sp. 1 and Bifidobacterium longum) were less frequent in C.
difficile ribotype 027 positive samples. Among them Bifidobacterium
longum had the highest statistical significance (P = 0.0031) (Table 3).
The analysis of the fungal groups showed that the diarrhoeal
samples positive for the C. difficile ribotype 027 were more
frequently colonised by Candida albicans and Candida glabrata,
compared to stools from healthy donors (Table 2). However, no
differences in fungi were detected among diarrhoeal samples
positive for the C. difficile ribotype 027 and other diarrhoeal routine
samples.
The machine learning approach identified four colonization
patterns associated with the 027 ribotype (Fig. 6). They are
different one to another and include bacteria, fungi and archaea.
The pattern with the best predictive score, albeit associated with
small number of samples (1 misclassified sample out of 5), involves
the presence of Enterobacteriaceae 1 and 2 as well as Candida glabrata.
In addition to the logistic regression analysis (Tables 2 and 3) and
the J48 model (Fig. 6), specific changes in the samples colonised by
the C. difficile ribotype 027 were detected also by the PCT model
(Table 5). Table 5 shows only those five microbial groups for

PLOS ONE | www.plosone.org 10 February 2013 | Volume 8 | Issue 2 | e58005


Figure 6. J48 decision tree describing the patterns of colonisation associated with presence or absence of C. difficile. Decision tree is
showing only routine sample sets. The decision tree shows combinations of microbial groups that predict the outcome regarding C. difficile
colonisation. The numbers in parentheses show the total number of samples and the number of incorrectly cassified instances, respectively.
doi:10.1371/journal.pone.0058005.g006
which the predicted outcomes between the sample sets were
different. All of them are predicted to be present in C. difficile
negative routine samples, but their absence is predicted in C.
difficile 027 positive samples.
Discussion
In this study we have used the DHPLC method to detect
changes in faecal microbial communities during C. difficile
colonisation. Although the method is simpler than more laborious
sequencing approaches and can detect only a small proportion of
present microbial groups, we have shown here that it does detect
differences between selected groups of faecal samples and that
detected differences are in agreement with studies published to
date.
The tested faecal samples were first distributed into those
collected from healthy volunteers and routine samples with
requested C. difficile testing. The latter were further divided into
six sets according to stool consistency, C. difficile presence or
absence and ribotype (027 or non-027) (Fig. 2 and 3).
In current study lower diversity was a common trait of routine
samples (formed or unformed) and not necessarily connected only
to C. difficile colonisation as average number of bacterial groups in
all routine samples was significantly lower compared to healthy
donors. Although 66 samples out of 171 routine samples were
negative for C. difficile, patients did have changed microbiota
(Fig. 5). The fact that samples were send to the routine testing
could be either due to diarrhoea (with C. difficile risk factors), while
for formed routine samples the reason for test requirement could
be untypical abdominal discomfort or retesting after CDI
treatment. No clinical data were collected to confirm this possible
explanation. Lower bacterial diversity was previously specifically
attributed to different gastrointestinal diseases. In addition to
PLOS ONE | www.plosone.org
Gut Microbiota Patterns and C. difficile
patients with a recurrent 8] or active CDI 11], lower bacterial
diversity was also observed in patients suffering from Crohns
disease (CD) 30], diarrhoea-predominant irritable bowel syn-
drome 31] and chemotherapy 32]. In addition, antibiotic
treatment has been shown to cause short-term 33] and long-term
34] shifts and decreased diversity in gut microbiota.
Our results are in concordance with the previous findings at the
level of individual bacterial taxons which showed lower levels of
Bifidobacterium longum, Prevotella sp. and Bacteroides sp.in C. difficile
positive samples 11,35,36]. In addition, in routine samples
Bifidobacterium longum proved to be the most important predictor
for the C. difficile negative status. Bacterial groups, which in our
study were more frequent in C. difficile positive samples, include
Ruminococcus bromii, the family Peptostreptococcaceae and Streptococcus
sp./Enterococcus sp. 2. The first two groups confirm the previous
studies which showed the increased numbers of firmicutes 36],
particularly Ruminococcus sp. in symptomatic adults 11] and
asymptomatic infants 37]; and increased levels of Clostridium sp.
in C. difficile colonised subjects 8,35]. In our study, significant
increase in colonisation frequencies of Streptococcus sp./Enterococcus
sp. 2 group were found in diarrhoeal C. difficile colonised samples.
Previous studies found a positive correlation between a persistent
asymptomatic C. difficile colonization and enterococci 38] and
connected vancomycin resistant enterococcus (VRE) infection to
more than 50% of C. difficile colonisations which prompted the
autors to suggest an active surveilance for VRE in C. difficile
colonised patients 39].
Very few studies have so far reported on the fungal microbiota
in connection to C. difficile colonization or the gut microbiota in
general. In C. difficile colonized hamsters treatment with mono-
clonal antibodies against C. difficile resulted in proliferation of
fungus Wickerhamomyces sp. 16]. In humans we have here described
11 February 2013 | Volume 8 | Issue 2 | e58005

several fungi to be found in healthy or disturbed faecal microbiota.
However, average number of fungal groups in faeces was low (1.5
2.0), which is similar to a previous study of fungal gut microbiota
in faeces of healthy individuals reporting 13 fungal species per
faecal sample 40]. Another culture independent study which
investigated fungal diversity in patients with inflammatory bowel
disease (IBD), showed a slightly higher number of mucosa-
associated fungi (6 per patient) 41]. Comparison of diversity in
patients with different types of intestinal inflammation (Crohns
disease (CD), ulcerative colitis, non-IBD colitis, unclear diarrhoea)
showed, that only CD patients had a higher fungal diversity 41].
We did not find differences in fungal diversity between all C.
difficile colonized and all non-colonized subjects. However, a
specific subgroup of diarrhoeal patients, colonised with the 027
ribotype, was more frequently colonised with Candida albicans and
Candida glabrata compared to the healthy donors. Our results
suggest that while fungal diversity does not increase during CDI,
there is a qualitative shift in the composition towards opportunistic
pathogens.
This is the first study to look at the possible changes in archaeal
microbiota in connection to C. difficile colonization. We found
higher frequencies of a single methanogenic species in the healthy
donors compared to C. difficile positive samples, but also when
comparing the C. difficile negative samples, sent for the routine C.
difficile testing, to the C. difficile positive samples. Our results are
therefore linking Methanobrevibacter smithii to the microbiota of
healthy individuals. Until recently, archaea have been a neglected
part of gut microbiota studies. The importance of archaea for the
homeostasis of the gut ecosystem and the role they play in the
bacterial degradation of sugars has led to the increased studies of
these microorganisms 42,43]. They have been implicated in
obesity after gastric bypass surgery 44] and in periodontal diseases
45].Because of the large number of samples colonised with the
ribotype 027, we could analyse them separately. Similar as
observed before on a much smaller set of samples 11], our results
also show that bacterial diversity in faecal samples colonized with
C. difficile 027 was lower than in other C. difficile positive samples.
Additionally, we showed that also certain fungi and archaea were
absent in faecal samples positive for C. difficile 027. Although so
called hypervirulence of ribotype 027 has to be considered with
caution 46], it seems that this particular strain has increased
potential to spread in the environment and to colonize the gut.
One of the possible explanations for this would be superior
colonizing properties of 027 strains and hence reduced need for an
References
1. Rupnik M, Wilcox MH, Gerding DN (2009) Clostridium difficile infection: new
developments in epidemiology and pathogenesis. Nat Rev Microbiol 7: 526536.
2. Johnson S, Samore MH, Farrow KA, Killgore GE, Tenover FC, et al. (1999)
Epidemics of diarrhea caused by a clindamycin-resistant strain of Clostridium
difficile in four hospitals. N Engl J Med 341: 16451651.
3. Graul T, Cain AM, Karpa KD (2009) Lactobacillus and bifidobacteria
combinations: a strategy to reduce hospital-acquired Clostridium difficile
diarrhea incidence and mortality. Med Hypotheses 73: 194198.
4. Mattila E, Uusitalo-Seppala R, Wuorela M, Lehtola L, Nurmi H, et al. (2012)
Fecal Transplantation, Through Colonoscopy, Is Effective Therapy for
Recurrent Clostridium difficile Infection. Gastroenterology 142: 490496.
5. Mellow MH, Kanatzar A (2011) Colonoscopic fecal bacteriotherapy in the
treatment of recurrent Clostridium difficile infection--results and follow-up.
J Okla State Med Assoc 104: 8991.
6. Ho N, Prasad V (2011) Clostridium difficile diarrhea and fecal transplantation.
J Clin Gastroenterol 45: 742743.
7. Buffie CG, Jarchum I, Equinda M, Lipuma L, Gobourne A, et al. (2012)
Profound alterations of intestinal microbiota following a single dose of
clindamycin results in sustained susceptibility to Clostridium difficile-induced
colitis. Infect Immun 80: 6273.
8. Chang JY, Antonopoulos DA, Kalra A, Tonelli A, Khalife WT, et al. (2008)
Decreased diversity of the fecal Microbiome in recurrent Clostridium difficile-
associated diarrhea. J Infect Dis 197: 435438.
PLOS ONE | www.plosone.org
Gut Microbiota Patterns and C. difficile
imbalanced microbiota. However, the association of ribotype 027
strains with the lowest microbiota diversity among all samples in
this study potentially suggest that ribotype 027 is capable of
actively affecting the composition of faecal microbiota and is
therefore increasing its own colonization potential.
Current studies have characterised C. difficile specific changes in
microbiota by evaluating relative frequencies and abundances for
each gut microorganism separately. Here, in contrast, we have
applied machine learning methods, and could show that C. difficile
colonisation is associated with patterns in the gut microbiota
composition. Some species are very strong predictors, as for
example, Bifidobacterium longum, which was connected to the C.
difficile negative outcome. Some other bacterial groups (e.g.
Prevotella) could be associated with a C. difficile positive or negative
status, which depends on the other microorganisms present in the
microbiota. This indicates that not only the presence or absence of
a single species/taxon but rather that certain combinations of gut
microbes are important in C. difficile associated changes of
microbiota.
Supporting Information
Table S1 The DHPLC separation conditions and peak
analysis parameters. Conditions for the separation of mixed
species 16S rDNA or ITS2 amplicons using the WAVE microbial
analysis system and peak analysis parameters for the Navigator
Software 2.2.0 (Build 25) are listed.
(DOCX)
Table S2 Average reciprocal Simpsons indexes of
diversity (1/D), standard deviations and P values of
the compared groups of samples.
(DOCX)
Acknowledgments
Authors would like to thank N. Oresic and I.Strumbelj for providing part
of the routine samples and to M. Medved for technical assistance.
Author Contributions
Conceived and designed the experiments: JS MR. Performed the
experiments: JS. Analyzed the data: JS SD BZ DM SG MR. Contributed
reagents/materials/analysis tools: MR SD. Wrote the paper: JS SD BZ
DM SG MR.
9. De La Cochetiere MF, Durand T, Lalande V, Petit JC, Potel G, et al. (2008)
Effect of antibiotic therapy on human fecal microbiota and the relation to the
development of Clostridium difficile. Microb Ecol 56: 395402.
10. Rousseau C, Lemee L, Le Monnier A, Poilane I, Pons JL, et al. (2011)
Prevalence and diversity of Clostridium difficile strains in infants. J Med
Microbiol 60: 11121118.
11. Rea MC, OSullivan O, Shanahan F, OToole PW, Stanton C, et al. (2012)
Clostridium difficile Carriage in Elderly Subjects and Associated Changes in the
Intestinal Microbiota. J Clin Microbiol 50: 867875.
12. Goldenberg O, Herrmann S, Marjoram G, Noyer-Weidner M, Hong G, et al.
(2007) Molecular monitoring of the intestinal flora by denaturing high
performance liquid chromatography. Journal of Microbiological Methods 68:
94105.
13. Wagner AO, Malin C, Illmer P (2009) Application of denaturing high-
performance liquid chromatography in microbial ecology: fermentor sludge,
compost, and soil community profiling. Appl Environ Microbiol 75: 956964.
14. Domann E, Hong G, Imirzalioglu C, Turschner S, Kuhle J, et al. (2003)
Culture-independent identification of pathogenic bacteria and polymicrobial
infections in the genitourinary tract of renal transplant recipients. Journal of
Clinical Microbiology 41: 55005510.
15. Rouge C, Goldenberg O, Ferraris L, Berger B, Rochat F, et al. (2010)
Investigation of the intestinal microbiota in preterm infants using different
methods. Anaerobe 16: 362370.
12 February 2013 | Volume 8 | Issue 2 | e58005

16. Peterfreund GL, Vandivier LE, Sinha R, Marozsan AJ, Olson WC, et al. (2012)
Succession in the Gut Microbiome following Antibiotic and Antibody Therapies
for Clostridium difficile. PLoS One 7: e46966.
17. Zidaric V, Zemljic M, Janezic S, Kocuvan A, Rupnik M (2008) High diversity of
Clostridium difficile genotypes isolated from a single poultry farm producing
replacement laying hens. Anaerobe 14: 325327.
18. Bidet P, Barbut F, Lalande V, Burghoffer B, Petit JC (1999) Development of a
new PCR-ribotyping method for Clostridium difficile based on ribosomal RNA
gene sequencing. FEMS Microbiol Lett 175: 261266.
19. Rupnik M, Avesani V, Janc M, von Eichel-Streiber C, Delmee M (1998) A novel
toxinotyping scheme and correlation of toxinotypes with serogroups of
Clostridium difficile isolates. J Clin Microbiol 36: 22402247.
20. Thompson JR, Marcelino LA, Polz MF (2002) Heteroduplexes in mixed-
template amplifications: formation, consequence and elimination by recondi-
tioning PCR. Nucleic Acids Res 30: 20832088.
21. Ovreas L, Forney L, Daae FL, Torsvik V (1997) Distribution of bacterioplankton
in meromictic Lake Saelenvannet, as determined by denaturing gradient gel
electrophoresis of PCR-amplified gene fragments coding for 16S rRNA. Appl
Environ Microbiol 63: 33673373.
22. Borneman J, Hartin RJ (2000) PCR primers that amplify fungal rRNA genes
from environmental samples. Appl Environ Microbiol 66: 43564360.
23. White TJ, Bruns T, Lee S, Taylor J (1990) Amplification and direct sequencing
of fungal ribosomal RNA genes for phylogenetics. In: PCR Protocols: a guide to
methods and applications Academic Press, New York, USA: 315322.
24. Turenne CY, Sanche SE, Hoban DJ, Karlowsky JA, Kabani AM (1999) Rapid
identification of fungi by using the ITS2 genetic region and an automated
fluorescent capillary electrophoresis system. J Clin Microbiol 37: 18461851.
25. Goldenberg O, Herrmann S, Adam T, Marjoram G, Hong G, et al. (2005) Use
of denaturing high-performance liquid chromatography for rapid detection and
identification of seven Candida species. J Clin Microbiol 43: 59125915.
26. Quinlan JR (1993) C4.5: Programs for Machine Learning. Morgan Kaufman,
San Francisco, CA, , USA.
27. Breiman L, Friedman J, Olshen R, Stone C (1984) Classification and Regression
Trees. Wadsworth, Belmont, CA.
28. Witten IH, Frank E, Hall MA (2011) Data mining: Practical Machine Learning
Tools and Techniques. Morgan Kaufman, San Francisco, CA, USA, third
edition.
29. Blockeel H, De Raedt L, Ramon J (1998) Top-down induction of clustering
trees. In: Proceedings of the Fifteenth International Conference on Machine
Learning (ICML 98) Morgan Kaufmann: 5563.
30. Manichanh C, Rigottier-Gois L, Bonnaud E, Gloux K, Pelletier E, et al. (2006)
Reduced diversity of faecal microbiota in Crohns disease revealed by a
metagenomic approach. Gut 55: 205211.
31. Carroll IM, Ringel-Kulka T, Siddle JP, Ringel Y (2012) Alterations in
composition and diversity of the intestinal microbiota in patients with
PLOS ONE | www.plosone.org
Gut Microbiota Patterns and C. difficile
diarrhea-predominant irritable bowel syndrome. Neurogastroenterol Motil 24:
521530, e248.
32. Zwielehner J, Lassl C, Hippe B, Pointner A, Switzeny OJ, et al. (2011) Changes
in human fecal microbiota due to chemotherapy analyzed by TaqMan-PCR,
454 sequencing and PCR-DGGE fingerprinting. PLoS One 6: e28654.
33. De La Cochetiere MF, Durand T, Lepage P, Bourreille A, Galmiche JP, et al.
(2005) Resilience of the dominant human fecal microbiota upon short-course
antibiotic challenge. J Clin Microbiol 43: 55885592.
34. Lofmark S, Jernberg C, Jansson JK, Edlund C (2006) Clindamycin-induced
enrichment and long-term persistence of resistant Bacteroides spp. and
resistance genes. J Antimicrob Chemother 58: 11601167.
35. Hopkins MJ, Macfarlane GT (2002) Changes in predominant bacterial
populations in human faeces with age and with Clostridium difficile infection.
J Med Microbiol 51: 448454.
36. Manges AR, Labbe A, Loo VG, Atherton JK, Behr MA, et al. (2010)
Comparative metagenomic study of alterations to the intestinal microbiota and
risk of nosocomial Clostridum difficile-associated disease. J Infect Dis 202: 1877
1884.
37. Rousseau C, Levenez F, Fouqueray C, Dore J, Collignon A, et al. (2011)
Clostridium difficile colonization in early infancy is accompanied by changes in
intestinal microbiota composition. J Clin Microbiol 49: 858865.
38. Ozaki E, Kato H, Kita H, Karasawa T, Maegawa T, et al. (2004) Clostridium
difficile colonization in healthy adults: transient colonization and correlation
with enterococcal colonization. J Med Microbiol 53: 167172.
39. Fujitani S, George WL, Morgan MA, Nichols S, Murthy AR (2011) Implications
for vancomycin-resistant Enterococcus colonization associated with Clostridium
difficile infections. Am J Infect Control 39: 188193.
40. Scanlan PD, Marchesi JR (2008) Micro-eukaryotic diversity of the human distal
gut microbiota: qualitative assessment using culture-dependent and -indepen-
dent analysis of faeces. Isme Journal 2: 11831193.
41. Ott SJ, Kuhbacher T, Musfeldt M, Rosenstiel P, Hellmig S, et al. (2008) Fungi
and inflammatory bowel diseases: Alterations of composition and diversity.
Scand J Gastroenterol 43: 831841.
42. Turnbaugh PJ, Ley RE, Mahowald MA, Magrini V, Mardis ER, et al. (2006) An
obesity-associated gut microbiome with increased capacity for energy harvest.
Nature 444: 10271031.
43. Samuel BS, Gordon JI (2006) A humanized gnotobiotic mouse model of host-
archaeal-bacterial mutualism. Proc Natl Acad Sci U S A 103: 1001110016.
44. Zhang H, DiBaise JK, Zuccolo A, Kudrna D, Braidotti M, et al. (2009) Human
gut microbiota in obesity and after gastric bypass. Proc Natl Acad Sci U S A 106:
23652370.
45. Eckburg PB, Lepp PW, Relman DA (2003) Archaea and their potential role in
human disease. Infect Immun 71: 591596.
46. Barbut F, Rupnik M (2012) Editorial commentary: 027, 078, and others: going
beyond the numbers (and away from the hypervirulence). Clin Infect Dis 55:
16691672.
13 February 2013 | Volume 8 | Issue 2 | e58005